Shu-Li Zhao
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Shu-Li Zhao has a total of 13 co-authors. Here are the top co-authors by number of shared publications:
- Ledong Chen
School of Electronic Science, National University of Defense Technology, Changsha, Hunan, China
5 shared publications - Jianfei Wu
College of Electronic Science and Technology, National University of Defense Technology, Changsha, Hunan, China
4 shared publications - Zhitao Xiao
Tiangong University
3 shared publications - Jiang Pingyun
National University of Defense Technology
2 shared publications - Jibin Liu
College of Electronic Science and Technology, National University of Defense Technology, Changsha, China
2 shared publications Caixu Yu
National University of Defense Technology
1 shared publication
Latest Publications
- 10 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Antennas and Propagation
- Vol. 74 (5) , 4021-4036
A Framework for Multiobjective Antenna Optimization Based on Differential Evolution of Deep Gaussian Processes: DGP–DEMO
In recent years, antenna design problems have become increasingly high-dimensional, strongly nonlinear, and tightly constrained. Existing surrogate-assisted frameworks for antenna optimization are limited in representational capacity, calibrated uncertainty, and handling of multiobjective trade-offs; consequently, they struggle to balance global exploration with precise convergence under restricted simulation budgets. To address these limitations, we propose DGP–DEMO, a surrogate-assisted multiobjective optimization framework for antenna design. DGP–DEMO employs a deep Gaussian process (DGP) as an online surrogate that continuously learns the mapping from design variables to electromagnetic responses and yields per-sample, calibrated uncertainty estimates; couples heuristic search with a Pareto-based fitness to efficiently generate and screen candidates across the parameter space; and introduces a surrogate-guided local refinement stage to improve feasibility and convergence quality. Case studies on two complex antennas show that, relative to existing surrogate-assisted frameworks, DGP–DEMO achieves effective multiobjective optimization of electromagnetic performance with fewer simulations and exhibits strong practical applicability.
- 10 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Antennas and Propagation
- Vol. 74 (5) , 4021-4036
A Framework for Multiobjective Antenna Optimization Based on Differential Evolution of Deep Gaussian Processes: DGP–DEMO
In recent years, antenna design problems have become increasingly high-dimensional, strongly nonlinear, and tightly constrained. Existing surrogate-assisted frameworks for antenna optimization are limited in representational capacity, calibrated uncertainty, and handling of multiobjective trade-offs; consequently, they struggle to balance global exploration with precise convergence under restricted simulation budgets. To address these limitations, we propose DGP–DEMO, a surrogate-assisted multiobjective optimization framework for antenna design. DGP–DEMO employs a deep Gaussian process (DGP) as an online surrogate that continuously learns the mapping from design variables to electromagnetic responses and yields per-sample, calibrated uncertainty estimates; couples heuristic search with a Pareto-based fitness to efficiently generate and screen candidates across the parameter space; and introduces a surrogate-guided local refinement stage to improve feasibility and convergence quality. Case studies on two complex antennas show that, relative to existing surrogate-assisted frameworks, DGP–DEMO achieves effective multiobjective optimization of electromagnetic performance with fewer simulations and exhibits strong practical applicability.
- 27 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Electromagnetic Compatibility
- Vol. 68 (2) , 404-413
Conducted Emission Behavior Model for System-in-Package Based on Multivariate Fitting
This article proposes and validates a novel approach to establish a behavior model for characterizing conducted emissions from system-in-package (SiP) chip. The method is based on data from S-parameter extraction and conducted emission measurements. By analyzing key parameters influencing the generation of conducted emissions, a typical correlation analysis is implemented to evaluate the correlation between the key parameters and noise sources. The mapping relationship between the main parameters and conductive emissions is then modeled using multivariate neural network fitting methods. Thereafter, a behavioral model reflecting the conductive emission characteristics of the SiP chip is constructed based on the ICEM-CE modeling standard. The method has been validated on real datasets of several functional modules in SiP chip, proving its effectiveness and accuracy. This method has a broad substantial application potential for evaluating electromagnetic compatibility (EMC) characteristics of systems and equipment.
- 27 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Electromagnetic Compatibility
- Vol. 68 (2) , 404-413
Conducted Emission Behavior Model for System-in-Package Based on Multivariate Fitting
This article proposes and validates a novel approach to establish a behavior model for characterizing conducted emissions from system-in-package (SiP) chip. The method is based on data from S-parameter extraction and conducted emission measurements. By analyzing key parameters influencing the generation of conducted emissions, a typical correlation analysis is implemented to evaluate the correlation between the key parameters and noise sources. The mapping relationship between the main parameters and conductive emissions is then modeled using multivariate neural network fitting methods. Thereafter, a behavioral model reflecting the conductive emission characteristics of the SiP chip is constructed based on the ICEM-CE modeling standard. The method has been validated on real datasets of several functional modules in SiP chip, proving its effectiveness and accuracy. This method has a broad substantial application potential for evaluating electromagnetic compatibility (EMC) characteristics of systems and equipment.
- Open Access
- 26 January 2026
- Published by MDPI AG in Electronics
- Vol. 15 (3) , 527
Diffusion Model Inverse Modeling and Applications to Microwave Filters
This paper presents a framework for inverse modeling of microwave filters based on a conditional diffusion model developed to address the intrinsic non-uniqueness of reconstructing coupling matrices from specified S-parameter responses. In the forward diffusion process, Gaussian noise is progressively added to the filter design variables, and a denoising network conditioned on the target electrical responses is trained to predict the injected noise at arbitrary diffusion steps. At inference, we initialize with Gaussian noise and execute the learned reverse denoising dynamics process; independent seeds yield diverse sets of physically feasible design-variable solutions that satisfy identical electrical-response constraints. Experiments on fourth- and sixth-order filters show that the proposed method outperforms multivalued neural networks (MVNNs) and conditional generative adversarial networks (CGANs) in prediction accuracy, solution diversity, and cumulative training cost, thereby providing a robust and efficient framework for inverse microwave-filter modeling.
- 26 January 2026
- Published by MDPI AG in Electronics
- Vol. 15 (3) , 527
Open AccessDiffusion Model Inverse Modeling and Applications to Microwave Filters
This paper presents a framework for inverse modeling of microwave filters based on a conditional diffusion model developed to address the intrinsic non-uniqueness of reconstructing coupling matrices from specified S-parameter responses. In the forward diffusion process, Gaussian noise is progressively added to the filter design variables, and a denoising network conditioned on the target electrical responses is trained to predict the injected noise at arbitrary diffusion steps. At inference, we initialize with Gaussian noise and execute the learned reverse denoising dynamics process; independent seeds yield diverse sets of physically feasible design-variable solutions that satisfy identical electrical-response constraints. Experiments on fourth- and sixth-order filters show that the proposed method outperforms multivalued neural networks (MVNNs) and conditional generative adversarial networks (CGANs) in prediction accuracy, solution diversity, and cumulative training cost, thereby providing a robust and efficient framework for inverse microwave-filter modeling.
- 2 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-10
MCU Function Recognition Method Based on the Frequency-Domain Constellation Trajectory Figure of Chip Electromagnetic Leakage and Convolutional Neural Networks
This paper introduces a novel chip electromagnetic fingerprint (EMF) feature extraction scheme based on the Frequency-domain Constellation Trajectory Figure (FCTF), as well as an EMF recognition method using Convolutional Neural Networks (CNN) for MCU function recognition. The IC-Stripline cell is a chip radiation emission measurement device, which has good shielding performance and field uniformity to clearly and completely receive chip electromagnetic (EM) leakage. The FCTF offers a two-dimensional visualization of the amplitude-frequency characteristics of chip electromagnetic leakage, effectively highlighting high-amplitude frequency features while suppressing low-amplitude ones. By employing CNNs to learn and recognize the distinct functional features of the FCTF, we have developed the FCTF-CNN chip EMF feature recognition approach. Compared with existing chip EMF recognition methods, the paper introduces several innovative contributions. In terms of data collection, we pioneer the application of the IC-Stripline cell for chip EMF acquisition which reduced testing complexity, marking a departure from conventional approaches. In terms of feature extraction, the proposed FCTF-CNN method not only achieves high recognition accuracy and low computational complexity but also exhibits unique robustness against Gaussian white noise interference. These advantages are thoroughly demonstrated through rigorous derivation and experimental validation. Experimental results show that under signal-to-noise ratios (SNR) of -20 dB and 0 dB, the FCTF-CNN method achieves recognition accuracies of 56.5% and 99.965% for the 14 functions of the STM32F103ZET6 MCU, respectively. These results are significantly superior to those of existing MCU EMF recognition methods.
- 2 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-10
MCU Function Recognition Method Based on the Frequency-Domain Constellation Trajectory Figure of Chip Electromagnetic Leakage and Convolutional Neural Networks
This paper introduces a novel chip electromagnetic fingerprint (EMF) feature extraction scheme based on the Frequency-domain Constellation Trajectory Figure (FCTF), as well as an EMF recognition method using Convolutional Neural Networks (CNN) for MCU function recognition. The IC-Stripline cell is a chip radiation emission measurement device, which has good shielding performance and field uniformity to clearly and completely receive chip electromagnetic (EM) leakage. The FCTF offers a two-dimensional visualization of the amplitude-frequency characteristics of chip electromagnetic leakage, effectively highlighting high-amplitude frequency features while suppressing low-amplitude ones. By employing CNNs to learn and recognize the distinct functional features of the FCTF, we have developed the FCTF-CNN chip EMF feature recognition approach. Compared with existing chip EMF recognition methods, the paper introduces several innovative contributions. In terms of data collection, we pioneer the application of the IC-Stripline cell for chip EMF acquisition which reduced testing complexity, marking a departure from conventional approaches. In terms of feature extraction, the proposed FCTF-CNN method not only achieves high recognition accuracy and low computational complexity but also exhibits unique robustness against Gaussian white noise interference. These advantages are thoroughly demonstrated through rigorous derivation and experimental validation. Experimental results show that under signal-to-noise ratios (SNR) of -20 dB and 0 dB, the FCTF-CNN method achieves recognition accuracies of 56.5% and 99.965% for the 14 functions of the STM32F103ZET6 MCU, respectively. These results are significantly superior to those of existing MCU EMF recognition methods.
- 24 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Antennas and Propagation
- Vol. 73 (6) , 3547-3559
A Deep-Learning-Based Modeling Method for Phase-Free Near-Field Scanning Measurement
This study presents a deep learning-based method for the precise inversion of electromagnetic radiation sources, utilizing amplitude data from multi-channel near-field scans to infer equivalent dipole array parameters. Unlike traditional heuristic search methods or interpolation-based artificial neural networks(ANN), the proposed model combines residual networks with channel attention mechanisms to effectively capture the intrinsic nonlinear relationships among multi-directional near-field data and their correlations with dipole parameters, thereby streamlining the inverse mapping process from near-field measurements to dipole parameters. Comprehensive simulations and experimental results, including scenarios with patch antennas at varying heights and MCU chips, validate the robustness and accuracy of the proposed model.
- 24 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Antennas and Propagation
- Vol. 73 (6) , 3547-3559
A Deep-Learning-Based Modeling Method for Phase-Free Near-Field Scanning Measurement
This study presents a deep learning-based method for the precise inversion of electromagnetic radiation sources, utilizing amplitude data from multi-channel near-field scans to infer equivalent dipole array parameters. Unlike traditional heuristic search methods or interpolation-based artificial neural networks(ANN), the proposed model combines residual networks with channel attention mechanisms to effectively capture the intrinsic nonlinear relationships among multi-directional near-field data and their correlations with dipole parameters, thereby streamlining the inverse mapping process from near-field measurements to dipole parameters. Comprehensive simulations and experimental results, including scenarios with patch antennas at varying heights and MCU chips, validate the robustness and accuracy of the proposed model.